A predictor model for the composting process on an industrial scale based on Markov processes

The biochemical and physical characteristics of composting processes have been historically modeled from an analytic point of view. Recently, stochastic approaches pushed forward the short-term forecast for the observed behaviour, but no model deals well with long-term predictions, especially when dealing with industrial data. We present a new approach, based on Markov processes, that shows good accuracy when predicting the long-term evolution of composting processes on an industrial scale. The proposed model deals with incomplete industrial data even for unevenly spaced observations, learns from past observations improving accuracy as data grows, and shows excellent predictive capabilities for time spans larger than 200 days and for heterogeneous large scale compost windrows. With our model, predictions can be obtained in real-time using Monte-Carlo runs. The model may be extremely convenient for industrial environments where large amounts of incomplete available data make it very difficult to use other prediction approaches.